2.6 Statistical analysis
For each habitat type and season, we decided the cumulative species richness and species abundance across all samplings and assembled the community matrix. Pollinator’s richness and abundance were compared between different habitats and seasons, using linear mixed–effect models with habitats and seasons as predictor variables and pollinators as response variables. The statistical analysis was performed in R, version 4.0.3. Tukey’s test was carried out to find significance.
PAST. Paleontological Statistics (Hammer et al., 2001) Version 3.17 computed the diversity indices. Random matrices with two samples are generated, each with the same row and column totals as in the original data matrix which provided the significance of diversity between groups.
Pollinator’s community compositions of different habitats (FT, GL, OT, MH) were analyzed by Non–metric Multidimensional Scaling (NMDS) of the abundance data employing the function meta MDS which is incorporated in the statistical package Vegan (Oksanen et al., 2013) and NMDS result with sample plots of different abundance scores was fitted with different habitats using the package ’ggplot2’ (Wickham, 2016).
NMDS was followed by statistical analyses: Adonis (Permutational Multivariate Analysis of variance), ANOSIM (Analysis of similarities), and SIMPER (Similarity Percentage Analysis).
Adonis was carried out following NMDS to analyze statistically if the pollinator community differs between the habitats. It provides the p–value to determine the statistical significance. ANOSIM, on the other hand, was used to determine if the differences of pollinator’s community between the habitats are significant. In addition to the significant difference tests, Simper analyses were used to identify those species that contributed most to the observed pollinator’s community differences (Clarke & Gorley, 2001).
To find relations between the environmental variables and the species composition, ordinations were performed on insect pollinators. For the pollinator community, a detrended correspondence analysis (DCA) was carried out to decide whether unimodal or linear ordination methods were appropriate (Lepˇs & ˇSmilauer, 2003). Based on this data, a redundancy analysis (RDA) was carried. Environmental variables were backward selected (p < 0.05) using the ‘ANOVA’. A Monte Carlo permutation test with 999 iterations was used to assess the significance of the ordination.
NMDS, RDA, and all of the three procedures (Adonis, ANOSIM, and SIMPER) were carried out in software R 4.0.3 using the ”vegan” package (Oksanen et al., 2013).
Venn diagrams showing the species sharing between the habitats were performed in R by using the “Venn Diagram” package employing the function draw. quad. venn.